Abstract:
Traditional measures of airline service quality provide useful indicators of performance, but they offer limited insight into why passengers are dissatisfied. This study uses a Large Language Model (LLM) to analyze unstructured passenger reviews and identify the specific service problems underlying changes in perceived service quality. We analyze 16,622 TripAdvisor reviews of two anonymized airlines, referred to as Airline A and Airline B, collected between 2016 and 2025 and covering 13 languages. The analysis classifies passenger complaints into 36 specific issues grouped into eight broader service categories. The results show sharply different rating trajectories: Airline A’s average rating fell from 3.3 in 2019 to 1.6 in 2024, while Airline B remained comparatively stable. The issue analysis provides a clearer explanation of these differences. Poor communication during delays is mentioned 536 times, compared with 690 mentions of flight delays and cancellations, while rude flight attendants are the most frequently reported customer-service complaint, with 591 mentions. The results also show substantial differences across passenger market groups. Overall, the findings suggest that broad service-quality measures can conceal the specific experiences that shape passenger evaluations. LLM-based analysis can complement conventional approaches by organizing large volumes of multilingual, unsolicited passenger feedback and showing which service problems appear most often and how they change over time.
